Propagate changes for tcSimplifyToDicts

Reduce renamer warnings

* Add Type.repType

* Re-express splitRepTyConApp_maybe using repType

* Use the new repType in Core2Stg

	The bug was that we ended up with a binding like
		let x = /\a -> 3# +# y
		in ...
	and this should turn into an STG case, but the big lambda
	fooled the core-to-STG pass

Improve common sub-expression stuff
	- better hash function
	- add Const.isBoxedDataCon, and use it in CSE
	- don't CSE for nullary constructors

Fix lost specialisations.  There were two problems

	{-# SPECIALISE f :: Int -> Rational #-}
	fromIntegral	=  fromInteger . toInteger

This generates

	fromIntegral_spec = fromIntegral d

for some suitable dictionary d.  But since fromIntegral is small,
it got inlined into fromIntegral_spec, thus losing the specialised
call (fromIntegral d) that was the whole raison d'etre of fromIntegral_spec.
Haskish solution: add an inlne pragma for the _spec things:

	fromIntegral_spec = _inline_me (fromIntegral d)

Now we won't inline inside.  But this showed up a related problem.  The
typechecker tries to common up overloaded things, so it actually generates

	m = fromIntegral d
	fromIntegral_spec = _inline_me m

which is pretty stupid.  Using tcSimplifyToDicts (instead of tcSimplify)
in TcBinds.tcSpecSigs fixes this.

a couple more weird cases for lexing of qualified names.

Fix some pretty subtle bugs in the lexing of qualified names.

writeHandle has been removed.

Move -fwarn-unused-{binds,matches} back into -W (two votes for, one
against).

nit-picking lexical tests for M.<reservedid> and M.<specialid>

Jump to the join point when returning a new constructor to a bind
default.  Fixes:  recent panic in mkStaticAlgReturnCode.

Don't treat GHC extension keywords as such when recognising qualified
varids.  This is a quick fix for the bug that M.label isn't recognised
as a qvarid with -fglasgow-exts on.

Don't attempt to close stdin or stdout.  From Kevin Atkinson
<kevinatk@home.com>.

Document the SocketPrim.PortNumber type

'do' expressions must have a non-empty list of statements.

Fix some race holes in the handle locking code, and clean it up a little.

Fix `defined but not used' warning to omit *all* identifiers beginning
with underscore, not just top-level ones, following Haskell report.

Rescue UsageSP analysis from bit-rot.

Resurrect -funfold-casms-in-hi-file and probably make Sigbjorn go wild
:-)

Readline needs Posix stuff

Make -optc-DDEBUG work again

Just a small test for my CVS setup: Changed two references to ghc-4.03
to ghc-4.04

+RTS -s gives summary statistics only.

Some fixes to the TICK_ macros in raiseAsync.

Set context_switch to 1 after scheduling a signal handler.  This will
ensure the running thread is interrupted as soon as possible.

Comment about TABLES_NEXT_TO_CODE.

New define: TABLES_NEXT_TO_CODE says whether info tables are assumed
to reside just before the code for a function.   We used to use
USE_MINIINTERPRETER for this function, but it makes sense in certain
cases to separate the use of tail-calls from tables-next-to-code.

StgMacros: add a couple of missing macros to support update in place.

Install the LICENSE file.

It's about time

Reinstate per-function entry counts in ticky-ticky profiles.

Increase stack for ParseIface a bit.

- Implement update-in-place in certain very specialised circumstances
- Clean up abstract C a bit
- Speed up pretty-printing absC a bit.

Switch on -fwarn-unused-binds and -fwarn-unused-matches by default; Sergeys suggestion

Fix non-compiling code

Some fixes to this (still non-working) pass.

The decision to not make a static closure should only be taken for
top-level bindings.

Minor cleanup

Nuke -fcase-elim; sometimes things fail if you dont have it (Sergeys message)

print minInt as (mintInt+1)-1, to avoid gcc's conversion to unsigned.

Bump version to 4.04 (to avoid confusion with already released 4.03
for Win32).